- Home
- Skills
- Workflow Automation
- Pointer: Drive Your Real Chrome
Works with the AI Agents you already use
Pointer: Drive Your Real Chrome
Let your agent click, type and read pages in your own Chrome, with your logins, while a visible pointer shows every move. Works with Claude Code, Cursor, Codex, OpenCode and more. Free and open source.
Free
Pointer: Drive Your Real Chrome
Example session with this skill installed
Use my browser to open Hacker News and take me to the comments of the most discussed story on the front page.
- Read your context and instructions
- Compiled the pointer
Summary
Opened Hacker News in a background tab of your Chrome, found the story with the most comments (483), and opened its discussion. You were never pulled away from your own tab; the agent's tab sits in a tab group named after the agent.
The session
These are the real commands and Pointer's real output (the MCP tools return the same text).
$ pointer open news.ycombinator.com
opened tab 1338347483 in the background: https://news.ycombinator.com/
$ pointer find "comments"
- link "comments" [ref=e5] href="/newcomments"
- link "41 comments" [ref=e17] href="/item?id=50268380"
- link "483 comments" [ref=e24] href="/item?id=50261319"
- link "22 comments" [ref=e31] href="/item?id=50234796"
- link "263 comments" [ref=e38] href="/item?id=50261998"
- link "116 comments" [ref=e45] href="/item?id=50264857"
- link "154 comments" [ref=e52] href="/item?id=50265609"
$ pointer click e24
clicked link "483 comments" → https://news.ycombinator.com/item?id=50261319
new page, current view:
url: https://news.ycombinator.com/item?id=50261319
title: agent · Genie 4 | Hacker News
view: 0% down · 0 elements above, 2891 below (scroll, find, or snapshot scope "all")
- link "Hacker News" [ref=e2]
- link "new" [ref=e3]
- link "comments" [ref=e5]
...
What happened on screen
| Step | What you saw in Chrome |
|---|---|
| open | A new tab appeared in the agent's tab group, in the background |
| find | Nothing moved: find reads the page and returns short refs |
| click e24 | The cursor slid to "483 comments" with a click · 483 comments label, rippled, and the discussion opened |
Every action reports only what changed, so the agent did not need another full snapshot after the click.
Result
The discussion of
Genie 4 (483 comments) is open in the agent's tab. Say pointer focus (or ask the agent) to bring it to the front, or pointer close when you are done; the agent never closes your own tabs.
Connects securely to your tools. The creator never sees your data.
About this skill
Pointer lets an AI coding agent use the Chrome you already have open: your profile, your cookies, your logged-in pages. Every action is drawn on screen. Before a click lands, a cursor slides to the target with a label like click · Sign Up, so you can watch what the agent does and stop it at any time.
What the skill does
This skill teaches your agent when and how to use Pointer safely:
- Snapshot, act, read the change. The agent lists what is on screen as short refs (
e3,e14), acts on a ref, and gets back only what changed. A six-step task uses about 2,400 tokens of tool output. - Real input. Clicks and keys are trusted browser events, so React apps, hover menus and pointer checks behave as they do for a person.
- Stays out of your way. The agent works in background tabs, grouped and labelled with its name, and never pulls you away from what you are doing.
- Safe by default. It confirms before purchases, sending messages, deleting or posting; never types passwords or card numbers; answers dialogs only with your yes; and stops when you press Esc or click in its tab.
- Debugging tools. Console errors, network requests, screenshots, responsive viewports and a layout checker for overflow and spacing bugs.
First use
Pointer has two free parts. On first use the skill walks your agent through setup:
- Install the command:
npm install -g @longvo2k/pointer(or the installer script). - Add the Pointer extension from the Chrome Web Store (one click, by you).
- Optional: register the MCP server so the agent gets Pointer's tools directly.
Good for
- Trying a signup or checkout flow and stopping before payment
- Checking a page that needs your login
- Filling forms and clicking through admin panels
- Debugging a UI with real console and network data
- Live demos, with on-screen captions for people watching
Open source
MIT licensed, no telemetry, and it only talks to 127.0.0.1. Source, issues and docs: https://github.com/longvo2k/pointer
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Workflow Automation.
Cannot drive chrome:// URLs or the Web Store. Does not traverse iframes/shadow roots. Requires manual extension installation. Single agent connection limit.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 20 days ago
- Free with an Agensi account
- Software
- Google ChromePointer extension
Node.js 18+@longvo2k/pointer
Both are free: the extension from the Chrome Web Store and the package from npm.